fix(notifications): workspace mark as read clears the chat alert on the phone - #33713

ELI5

When an agent chat finishes, Orca can send an alert to your paired phone. Clicking "Mark as read" on the workspace row on the desktop cleared the dot, but the phone alert stayed. Now Mark as read clears the phone alert too, the same way opening the chat or marking it read in Activity already does.

What Changed

• Before: workspace Mark as read (the row's quick action and the context menu) removed the workspace's unread markers and asked main to dismiss notifications by pane key only. Chat completion alerts are tied to a position in the session's journal, and main only retires those when it receives a read that covers that position. With no such read, the phone alert stayed in the tray.
• After: setWorkspaceUnreadByUser also emits an explicit subject read (emitAgentSubjectReads(..., 'explicit')) for every pane whose marker it cleared. The existing chat attention bridge turns that into a dismiss carrying the journal boundary (the edge that lit the row, never a later one), plus the host acknowledgement. This retires:
  • the alert desktop main relayed to its phones for a chat on a remote paired host;
  • the alert a local host pushed to its phones directly.
• Terminal agents are unchanged. They have no read capture, so the extra read is ignored for them.

Why

Root cause: the workspace toggle was the only read path that never sent a structured read. Opening the chat (the on-screen auto-ack) and Activity/dashboard Mark as read (acknowledgeAgents(..., 'explicit')) both go through emitAgentSubjectReads. This reuses that same path instead of adding anything new.

I didn't call acknowledgeAgents from the workspace toggle. It also writes Activity acknowledgement stamps and clears manual-unread turns, so marking a workspace read would start marking its Activity threads read. That's a separate product call.

No overlap with #30900: that PR doesn't touch the slice or the subject-read capture/listener path.

Linked Issue

Linear STA-11936 (no GitHub issue).

Visual Proof

N/A: nothing changes on the desktop. The change is that the phone's tray alert is withdrawn, which needs a paired host and a phone running a real chat turn. The regression test covers the emitted read, and the downstream dismissal is the same chain Activity's Mark as read already uses.

Testing

• Added a test showing workspace Mark as read emits an explicit read for a cleared chat subject. It fails on main (expected [] to deeply equal ['explicit']) and passes with the fix.

• Ran pnpm test src/renderer/src/store/slices/workspace-attention src/renderer/src/hooks/useReadOnScreenAttention.test.ts, check:code-quality:changed and check:react-doctor:changed.

• Not run: a live phone tray check, for the reasons under Visual Proof.

• I manually tested these changes locally

• Automated tests added/updated, or explained why not below

AI Disclosure

Review

Renderer-only. One call added in workspace-attention-slice.ts; removeAndDismiss now returns what it removed instead of a boolean.

Agent skill upstream boundary

• Not applicable, or this change copies or mechanically translates no upstream skill-installer source, tests, fixtures, registry entries, path tables, comments, or documentation.

Notes

• Platforms: renderer-only, so no platform differences.
• Remote/SSH: remote-host chats are the case this fixes. The host acknowledgement RPC is the one other read paths already send, so mixed versions behave as they do for those paths.
• Folder workspaces: this path doesn't depend on git, so they work the same.
• Mobile: no phone change needed. The phone already handles these dismissals.

Still not covered (out of scope, pre-existing):

• alerts from before a desktop restart, when no unread markers remain;
• pushes sent by a headless host;
• pushes that carry no notification id.

Checklist

• This PR is small and focused
• I explained what changed and why (ELI5, the user-facing before/after, the mechanism, and why over the alternatives)
• Before/after screenshots or videos attached for UI changes, or N/A with reason
• Self-reviewed for correctness, security, and performance
• Cross-platform, SSH/remote, and path/shortcut impact considered (or N/A)
• pnpm lint, pnpm typecheck, pnpm test, and pnpm build pass (or CI will cover; local preferred)